For dataset preparation and preprocessing, we assume you already installed the conda environment and finished installation of SMPL model. For data preprocessing, you also need to download this .npy file from the EasyMocap Repository, and put it under body_models/misc/
To download the ZJU-MoCap dataset please get access here. After you get the dataset, extract the dataset to an arbitrary directory, denoted as ${ZJU_ROOT}. It should have the following structure:
${ZJU_ROOT}
├-- CoreView_313
├-- CoreView_315
| ...
└-- CoreView_394
To preprocess one sequence (e.g. CoreView_313), run the following
export PYTHONPATH=${PWD} # only need to run this line once
python preprocess_datasets/preprocess_ZJU-MoCap.py --data-dir ${ZJU_ROOT} --out-dir ${OUTPUT_DIR} --seqname CoreView_313
where ${OUTPUT_DIR} is the directory where you want to save the preprocessed data. After this, create a symbolic link under ./data directory by:
ln -s ${OUTPUT_DIR} data/zju_mocap
Our H36M models are trained with the preprocessed version of H36M data from Animatable NeRF. Please first email the authors of Animatable NeRF to get a copy of their preprocessed version of H36M data.
After you get the data, extract the dataset to an arbitrary directory, denoted as ${H36M_ROOT}. It should have the following structure:
${H36M_ROOT}
├-- S1
├-- S2
| ...
└-- S11
To preprocess one sequence (e.g. S9), run the following
export PYTHONPATH=${PWD} # only need to run this line once
python preprocess_datasets/preprocess_H36M.py --data-dir ${H36M_ROOT} --out-dir ${OUTPUT_DIR} --seqname S9
where ${OUTPUT_DIR} is the directory where you want to save the preprocessed data. After this, create a symbolic link under ./data directory by:
ln -s ${OUTPUT_DIR} data/h36m
Note that for preprocessing AIST++, we assume you already setup ZJU-MoCap dataset as instructed, and a symbolic link to the dataset exists as ./data/zju_mocap. You only need motion data from AIST++, please first get the motion data from the official website.
After you get the data, extract the dataset to an arbitrary directory, denoted as ${AIST_ROOT}. It should have the following structure:
${AIST_ROOT}
└-- motions
To transfer poses of sequence gBR_sBM_cAll_d04_mBR1_ch06 of AIST++ to subject CoreView_377 of ZJU-MoCap, run the following
export PYTHONPATH=${PWD} # only need to run this line once
python preprocess_datasets/preprocess_aist.py --data-dir ${AIST_ROOT}/motions --seqname gBR_sBM_cAll_d04_mBR1_ch06 --subject CoreView_377
by default the data will be saved to respective directories under ./data/odp.
Coming soon!